The problem
Most small businesses — clinics, beauty salons, private practices — still coordinate appointments over WhatsApp manually: the owner receives a message, checks a paper calendar or the phone’s native calendar app, and replies with availability. The process is slow, prone to double-booking errors, and doesn’t scale when multiple employees or locations are involved. The challenge was to automate that workflow without forcing the end customer to switch channels: WhatsApp is already where they are.
The solution
Agenda IA is a multi-tenant SaaS where each subscribed business gets a dedicated WhatsApp number. When a customer sends a message like “I’d like an appointment on Monday at 10 for a haircut”, the system captures it through the WhatsApp Cloud API, forwards it to Claude AI for intent extraction and entity recognition (date, time, service type), validates availability against Google Calendar, and confirms the booking by responding in natural language within the same conversation thread — all in under three seconds.
Architecture
The backend runs on Node.js with Express. Redis handles the incoming message queue to absorb traffic spikes without dropping webhooks, and Socket.IO pushes real-time events to the admin dashboard — built with React, Vite, and TypeScript — where business owners see confirmed, cancelled, and pending appointments without refreshing the page. The Google Calendar API integration is bidirectional: if an appointment is cancelled from the dashboard, the event is removed from the calendar and the bot notifies the customer over WhatsApp. MySQL stores the conversation history, per-tenant configuration, and availability rules (schedules, services, duration per appointment type). Multi-tenancy is implemented at the database level with a tenant_id column on all critical tables, and incoming WhatsApp webhooks are routed by phone number to the correct tenant context.
Key decisions
Three product decisions shaped the system more than the stack itself:
- Stay inside WhatsApp instead of building a new booking channel. The goal was not to teach customers a new interface; it was to automate the one they already use every day.
- Respond fast, process later. Webhook reception had to be decoupled from AI and calendar logic so the system could acknowledge messages immediately and finish the heavy work asynchronously.
- Treat each business as its own operational context. Queues, configuration, schedules and conversations all had to respect tenant boundaries so one busy client would not degrade the rest.
Product constraints
Agenda IA was not only a chatbot problem. It had to solve for operational ambiguity:
- People write requests in messy natural language
- Businesses use different schedules, services and durations
- Two customers can ask for the same slot at nearly the same time
- Owners need to trust the result without supervising every message manually
That forced the product to be conservative. The system does not try to sound magical first; it tries to be correct first, then natural.
Outcome
The result is a booking flow that feels conversational to the customer and structured to the business owner. On one side, the user sends a WhatsApp message in plain language and receives a confirmation in the same thread. On the other, the business gets a synchronized operational layer: real-time admin updates, calendar consistency, centralized tenant configuration and a queue-backed message pipeline that keeps the system responsive under load.
The strongest part of the project is not the AI label. It is the way interface, infrastructure and operations were made to behave like one product.
